Amenity, Diversity and Obesity: Unobserved Heretogeneity in Cities

32 Pages·2015·20.25 MB·English

Institute for International Economic Policy Working Paper Series
Elliott School of International Affairs
The George Washington University
Amenity, Diversity and Obesity: Unobserved Heretogeneity in
Cities
IIEP-WP-2015-12
Stephen Popick
Anthony Yezer
George Washington University
June 2015
Institute for International Economic Policy
1957 E St. NW, Suite 502
Voice: [phone]
Fax: [phone]
Email: [email protected]
Web: www.gwu.edu/~iiep
AMENITY, DIVERSITY AND OBESITY: UNOBSERVED
HERETOGENEITY IN CITIES
STEPHEN POPICK#, AND ANTHONY M. YEZER+
June, 2015
ABSTRACT. Some sources of heterogeneity among cities, i.e. age, gender, race,
income, and education, have been the object of substantial inquiry. The reasons
are obvious. These differences are easily observed and may have important
implications for economic activity. This study considers another potentially
important population characteristic, obesity. Descriptive statistics reveal that the
intercity variance in obesity rates is substantial. Empirical results demonstrate
that demographic and regional amenity variables all have a relation to intercity
differences in obesity. Because obesity is important for preferences, performance,
and productivity, its omission from previous studies and its correlation with
amenity and demographic characteristics, could create problems for empirical
research.
JEL Codes: I12, J10, R23,
_________________________________________________________________________
# Ph.D. Candidate, Department of Economics, George Washington University 505 A E Windsor
Ave Alexandria VA 22301 [phone] <[email protected]>
+ Corresponding Author, Professor, Department of Economics, George Washington University,
2100 G Street NW Washington DC 20006 [phone] <[email protected]>
The authors declare that they have no relevant or material financial interests that relate to the
research described in this paper. None of the data are proprietary.
AMENITY, DIVERSITY AND OBESITY: UNOBSERVED
HERETOGENEITY IN CITIES
I. INTRODUCTION
It is well known that obesity is negatively associated with income and education. 1 Age,
ethnicity, race, and gender effects are more complex but certainly significant. Because cities
differ in composition along income, education, age, race, ethnicity, and gender dimensions, they
should naturally differ in obesity rates. However, a simple comparison of obesity rates across
cities suggests that other factors may be influential. Approximately 42.7% of the population of
the San Francisco-Oakland, CA MSA have a non-obese Body Mass Index (BMI < 25) while
20.7% are obese (BMI > 30). In contrast, the Detroit, MI MSA percentages are 29.2% non-obese
and 35.6% obese. Could such differences in obesity rates be due to observable age, race,
ethnicity, gender, education, and income characteristics of individuals in the population of these
cities or are specific city characteristics differentially attractive to the obese?
The question addressed in this paper is whether, in addition to observable personal
characteristics of their inhabitants, city amenities are associated with differences in obesity. The
literature reviewed here suggests a number of reasons to believe that differences in climate,
topography, and of course food prices, make cities differentially attractive to individuals with
1" Much of this knowledge is based on the quesons in the Naonal Longitudinal Survey of Youth. This
sample is inadequate to test the hypotheses regarding intercity di#erences being examined here. It is
quite adequate to demonstrate di#erences in BMI associated with personal characteriscs. The other
source of BMI data is the Naonal Health Inventory Survey, which is annual, but only iden)es 33 MSAs.
This number is too small for the analysis of the e#ects of di#erences in city characteriscs undertaken
here.
high BMI. This differential attraction may then cause individuals to select into these cities
through migration. To the extent that differences in BMI are hereditary and city amenities are
permanent, past migration may also select those prone to obesity into certain cities. Thus, there
may be a “BMI selection effect” of city characteristics. 2 Alternatively, there may be a “BMI
adaptation effect” as individuals adjust their BMI to city characteristics. Regardless of the
process, both selection and adaptation effects, if they are important, may contribute significantly
to the large spatial differences in BMI documented in this paper.3
Why is the possibility that spatial characteristics play a role in determining BMI
differences in the resident population important? First, general interest in obesity is high
because of its strong connection with development of type 2 diabetes mellitus (DM) and other
ailments that impose substantial costs on society. Recent estimates of these costs are as large as
1.1 trillion dollars per year for the U.S. economy. 4 Second, there is an important issue for
empirical research in economics. BMI is generally unobservable to the econometrician.
Nevertheless, BMI is an important determinant of individual preferences and worker productivity
partly due to direct effects and indirectly through the connection between obesity and DM.
Therefore, because BMI is correlated with variables such as income, education, and demographic
2" The BMI selecon e#ect has created problems in the literature on the e#ects of “sprawl” on BMI.
There is substanal disagreement in this literature. See, for example, Eid, Overman, Puga, and Turner
(2008) versus Zhao and Kaestner (2010). This paper is not concerned with the distribuon of BMI within
cies.
3" It may be that individual expectaons for “opmal” BMI are based on community standards and that
behavioral economics could explain local variaon in diet and exercise. Following Brennan’s (2014)
recent suggeson, the analysis can be viewed as tesng a raonal choice model.
4"Brookings Instuon Study. See: h<p://www.brookings.edu/blogs/brookings-
now/posts/2015/05/societal-costs-of-obesity
characteristics that are important in empirical research, this raises the possibility for omitted
variable bias in estimates of the effects of these personal characteristics on measures of wage
differentials for constant quality workers, and in a variety of other empirical research.
Alternatively, if BMI is related to city amenity characteristics, this unobserved association could
confound inferences about causes of spatial wage or productivity differences across cities. 5
Could it be that a significant portion of the wage differential between observationally similar
workers in San Francisco and Detroit is due to unobserved differences in their BMIs?6
Recent availability of large scale individual survey data on BMI for a representative
sample of city populations allows testing of the hypothesis that, holding income, education, and
demographic characteristics constant, selected city characteristics have a significant relation to
their obesity rates.7 The object of this study is to test the hypothesis that the city characteristics
that are expected, based on physiological effects of obesity, to make areas differentially attractive
to those with high BMI, have an influence on the average body mass index and obesity rate in
the city.
The next section of this paper develops the theoretical rational for believing that there is a
BMI selection effect in which city characteristics have a differential attraction for obese
5" There is a substanal quality of life following Roback (1982) that relates wage di#erenals to city
amenies under the assumpon that amenies have no e#ect on obesity or other unobservable
populaon characteriscs.
6" Observaonal equivalence in this case refers to research that does not observe worker body mass
index (BMI).
7"The Centers for Disease Control Behavioral Risk Factor Surveillance System surveys of individual BMI
used in this study have been conducted for many years but, over the past 15 years the sample size
increased signi)cantly so that reliable esmates of BMI di#erences across a range of cies are possible.
individuals. Then the available literature that relates BMI to preferences for climate, topography,
and other city characteristics is reviewed. The data section discusses the construction of
variables designed to measure these city differences. Finally, empirical results show general
agreement between prior expectations and the obesity rate of cities.
II. THEORY: BMI AND CHOICE OF LOCATION
Assume that there are multiple households differentiated by a single scalar characteristic,
B, which is an “inherited” property of individuals. 8 They must choose a location among areas
indexed by j that are differentiated by wages, w, transportable goods, x, whose price everywhere
j
is p, non-transportable goods, h, for “housing” whose price, r, varies spatially, and local amenity
j
whose implicit price, q, varies spatially. The indirect utility of a particular household , i, in
j
location j can be written as:
U = V(w, p, r, q; B) (1)
ij j j j i
Taking the total differential of indirect utility under the assumption of constant utility
across cities and solving for dw gives:
j,
dw =−λ dp−λ dr −λ dq
j xw hw j qw j (2)
8" For purposes of this model, it does not ma<er whether B has a genec origin or if it is learned in
childhood.
λ
where yz is the ratio of the partial derivative V(.) with respect to y divided by the partial of
V(.) with respect to z.9 Applying Roy’s identity, the relation in equation (2) can be solved for the
total derivative of earnings:
dw =x dp+h dr −a dq
j j j j j j (3)
which, assuming dp = 0, implies that:
dw =h dr −a dq
j j j j j
(4) Equation (4) states that the equilibrium tradeoff between wages and rents depends
on the quantity of amenity consumed by the individual. It follows from the effect of B on
indirect utility of the amenity that da/dB > 0 and high B households will require a smaller
j
compensating differential in wages to live in areas where the price of the amenity in question is
lower. Therefore, high B households are differentially attracted into areas with low q.
j
Obviously, the B factor relevant for this research is BMI and the hypothesis is that the
relative concentration of high BMI individuals will rise in areas where the prices of amenity
factors that are differentially attractive to obese are low. This spatial sorting of population by
BMI could arise through migration and/or genetic selection. Ford (2005) has noted that migration
is one possible sorting mechanism under the hypothesis that the tendency to be obese varies
significantly in the population. Alternatively, Piziak (2010) and Andersson (2011) contend that
heredity is an important determinant of BMI. This suggests that spatial differences in BMI could
be the result of prior migration by those with a genetic predisposition to obesity. Finally, Chen
9" Note that (1) can be wri<en in implicit form and provide a clear statement of the spaal iso-ulity
condion.
(2013) has observed that standards of diet and exercise could vary spatially based on the
interaction of preferences, which vary with BMI, and the relative proportion of the obese in the
population.
This paper does not test the exact mechanism that accomplishes the sorting, although
estimation results indicate that the effects of climate and topography on BMI for individuals
under 25 years of age are identical to those for individuals 25 or older. To the extent that
migration occurs at ages greater than 24, this suggests that differences in BMI are not due to
recent migration but rather to differences in the resident populations of areas that could be the
result of past migration.10
III. EFFECTS OF INDIVIDUAL AND CITY AMENITY
CHARACTERISTICS ON OBESITY
The four most prominent individual characteristics recognized in the economics literature
as having a possible relation to BMI are income, education, gender, and age. The empirical
evidence strongly suggests that obesity varies inversely with both income and education (Baum
2004). The underlying reasons for this relation are not clear and may be quite complex but the
relation holds within as well as among cities. Females have lower BMI. 11 Age effects are non-
linear because there is a tendency for BMI to be highest in middle age (Gallup 2012). To the
extent that income, education, gender, and age are distributed unequally across cities, they may
explain a significant portion of the variation in spatial obesity rates. In addition, race and
10" Limited sample size and the discrete categories of age did not permit tesng of BMI e#ects for
younger age cohorts.
11" CDC/NCHS, Health, United States, 2014, Table 64. Data from the Naonal Health and Nutrion
Examinaon Survey (NHANES)
ethnicity are unequally distributed across cities and may have an independent relation to BMI.
The purpose of this research is not to sort out the causal relation between these factors and BMI
but rather to test whether spatial differences in their distribution can explain the large differences
in BMI and obesity across cities or if other factors involving population selection based on
amenities analyzed in the theory section are important.
Much of the literature on amenity factors, whose attractiveness might vary with
individual BMI, lies outside of economics because physiology is the basis for differences in
preferences. Simply put, endomorphs react differently than ectomorphs to the same
environmental conditions. Examination of the literature reveals a number of area characteristics
that should relate to obesity because the preferences of the obese are observed to differ from the
average and thin populations. These preferences are the result of physiological effects of obesity.
A substantial literature, stemming from seminal work by Roback (1982), classifies these factors
as local amenities. First are opportunities for outdoor recreation. City characteristics including
access to water and parkland should be valued less by the obese. Secondly, the obese have
difficulty dealing with certain topographic characteristics. Voss (2013) has found that elevation
and elevation change are more physically demanding for the obese. Accordingly, individuals
with high BMI will avoid mountainous locations and seek relatively flat coastal locations
because of both topography and oxygen availability. BMI has a significant effect on preferences
for climate. Cold winters are more uncomfortable for those with low BMI. Hot summers make
outdoor recreation more difficult. Lin (2007) has argued that, given that exercise is associated
with lower BMI, individuals who exercise try to avoid areas where summers are extremely hot.
Dehghan et. al. (2013) found obese workers suffered more cardiac strain than their non-obese
colleagues did in hot and humid conditions. Accordingly, those with low BMI have a relatively
stronger preference for areas with mild winters and mild summers. Causality can go in either
direction here. Cold winters and hot summers may make outdoor exercising difficult and the
lack of exercise may lead to higher BMI. As noted above, for purposes of this study, the
direction of causality is not material.
One limitation of the explicit measures of environmental amenity discussed above is that
measures of parkland, water bodies, etc., are not adjusted for quality of the recreational
experience that they provide. To the extent that the quality dimension of these local amenity
variables is missing, there is measurement error that causes attenuation bias in estimates of the
amenity effect.
Research on obesity has isolated a number of other non-amenity factors that tend to repel
the obese and/or attract ectomorphs. Edwards (2008) identified the availability or use of public
mass transit, Booth (2005) cited housing density, and Grossman (2013) among others pointed to
the cost of food as factors that may relate to obesity rates in an area. These additional factors are
added to the empirical analysis.
IV. DATA ON BMI AND URBAN AMENITY FACTORS
In relating cross section variation in body mass index (BMI) to the city characteristics
identified above, availability of data has previously been a major constraint. The source of BMI
data for this study is the 2010 Centers for Disease Control (CDC) Behavioral Risk Factor
Surveillance System survey of BMI at the individual and 2010 CDC SMART data (derived from